1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Linux INET6 implementation
9 * Based on linux/ipv4/udp.c
12 * Hideaki YOSHIFUJI : sin6_scope_id support
13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
15 * a single port at the same time.
16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
20 #include <linux/bpf-cgroup.h>
21 #include <linux/errno.h>
22 #include <linux/types.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/net.h>
26 #include <linux/in6.h>
27 #include <linux/netdevice.h>
28 #include <linux/if_arp.h>
29 #include <linux/ipv6.h>
30 #include <linux/icmpv6.h>
31 #include <linux/init.h>
32 #include <linux/module.h>
33 #include <linux/skbuff.h>
34 #include <linux/slab.h>
35 #include <linux/uaccess.h>
36 #include <linux/indirect_call_wrapper.h>
38 #include <net/addrconf.h>
39 #include <net/ndisc.h>
40 #include <net/protocol.h>
41 #include <net/transp_v6.h>
42 #include <net/ip6_route.h>
45 #include <net/tcp_states.h>
46 #include <net/ip6_checksum.h>
47 #include <net/ip6_tunnel.h>
49 #include <net/inet_hashtables.h>
50 #include <net/inet6_hashtables.h>
51 #include <net/busy_poll.h>
52 #include <net/sock_reuseport.h>
54 #include <linux/proc_fs.h>
55 #include <linux/seq_file.h>
56 #include <trace/events/skb.h>
59 static void udpv6_destruct_sock(struct sock *sk)
61 udp_destruct_common(sk);
62 inet6_sock_destruct(sk);
65 int udpv6_init_sock(struct sock *sk)
67 skb_queue_head_init(&udp_sk(sk)->reader_queue);
68 sk->sk_destruct = udpv6_destruct_sock;
72 static u32 udp6_ehashfn(const struct net *net,
73 const struct in6_addr *laddr,
75 const struct in6_addr *faddr,
78 static u32 udp6_ehash_secret __read_mostly;
79 static u32 udp_ipv6_hash_secret __read_mostly;
83 net_get_random_once(&udp6_ehash_secret,
84 sizeof(udp6_ehash_secret));
85 net_get_random_once(&udp_ipv6_hash_secret,
86 sizeof(udp_ipv6_hash_secret));
88 lhash = (__force u32)laddr->s6_addr32[3];
89 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
91 return __inet6_ehashfn(lhash, lport, fhash, fport,
92 udp_ipv6_hash_secret + net_hash_mix(net));
95 int udp_v6_get_port(struct sock *sk, unsigned short snum)
97 unsigned int hash2_nulladdr =
98 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
99 unsigned int hash2_partial =
100 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
102 /* precompute partial secondary hash */
103 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
104 return udp_lib_get_port(sk, snum, hash2_nulladdr);
107 void udp_v6_rehash(struct sock *sk)
109 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
110 &sk->sk_v6_rcv_saddr,
111 inet_sk(sk)->inet_num);
113 udp_lib_rehash(sk, new_hash);
116 static int compute_score(struct sock *sk, struct net *net,
117 const struct in6_addr *saddr, __be16 sport,
118 const struct in6_addr *daddr, unsigned short hnum,
121 int bound_dev_if, score;
122 struct inet_sock *inet;
125 if (!net_eq(sock_net(sk), net) ||
126 udp_sk(sk)->udp_port_hash != hnum ||
127 sk->sk_family != PF_INET6)
130 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
136 if (inet->inet_dport) {
137 if (inet->inet_dport != sport)
142 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
143 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
148 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
149 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif);
155 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
161 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
163 const struct in6_addr *saddr,
165 const struct in6_addr *daddr,
168 struct sock *reuse_sk = NULL;
171 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
172 hash = udp6_ehashfn(net, daddr, hnum, saddr, sport);
173 reuse_sk = reuseport_select_sock(sk, hash, skb,
174 sizeof(struct udphdr));
179 /* called with rcu_read_lock() */
180 static struct sock *udp6_lib_lookup2(struct net *net,
181 const struct in6_addr *saddr, __be16 sport,
182 const struct in6_addr *daddr, unsigned int hnum,
183 int dif, int sdif, struct udp_hslot *hslot2,
186 struct sock *sk, *result;
191 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
192 score = compute_score(sk, net, saddr, sport,
193 daddr, hnum, dif, sdif);
194 if (score > badness) {
195 result = lookup_reuseport(net, sk, skb,
196 saddr, sport, daddr, hnum);
197 /* Fall back to scoring if group has connections */
198 if (result && !reuseport_has_conns(sk, false))
201 result = result ? : sk;
208 static inline struct sock *udp6_lookup_run_bpf(struct net *net,
209 struct udp_table *udptable,
211 const struct in6_addr *saddr,
213 const struct in6_addr *daddr,
214 u16 hnum, const int dif)
216 struct sock *sk, *reuse_sk;
219 if (udptable != &udp_table)
220 return NULL; /* only UDP is supported */
222 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_UDP, saddr, sport,
223 daddr, hnum, dif, &sk);
224 if (no_reuseport || IS_ERR_OR_NULL(sk))
227 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
233 /* rcu_read_lock() must be held */
234 struct sock *__udp6_lib_lookup(struct net *net,
235 const struct in6_addr *saddr, __be16 sport,
236 const struct in6_addr *daddr, __be16 dport,
237 int dif, int sdif, struct udp_table *udptable,
240 unsigned short hnum = ntohs(dport);
241 unsigned int hash2, slot2;
242 struct udp_hslot *hslot2;
243 struct sock *result, *sk;
245 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
246 slot2 = hash2 & udptable->mask;
247 hslot2 = &udptable->hash2[slot2];
249 /* Lookup connected or non-wildcard sockets */
250 result = udp6_lib_lookup2(net, saddr, sport,
251 daddr, hnum, dif, sdif,
253 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
256 /* Lookup redirect from BPF */
257 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
258 sk = udp6_lookup_run_bpf(net, udptable, skb,
259 saddr, sport, daddr, hnum, dif);
266 /* Got non-wildcard socket or error on first lookup */
270 /* Lookup wildcard sockets */
271 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
272 slot2 = hash2 & udptable->mask;
273 hslot2 = &udptable->hash2[slot2];
275 result = udp6_lib_lookup2(net, saddr, sport,
276 &in6addr_any, hnum, dif, sdif,
283 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
285 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
286 __be16 sport, __be16 dport,
287 struct udp_table *udptable)
289 const struct ipv6hdr *iph = ipv6_hdr(skb);
291 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
292 &iph->daddr, dport, inet6_iif(skb),
293 inet6_sdif(skb), udptable, skb);
296 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
297 __be16 sport, __be16 dport)
299 const struct ipv6hdr *iph = ipv6_hdr(skb);
301 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
302 &iph->daddr, dport, inet6_iif(skb),
303 inet6_sdif(skb), &udp_table, NULL);
306 /* Must be called under rcu_read_lock().
307 * Does increment socket refcount.
309 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
310 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
311 const struct in6_addr *daddr, __be16 dport, int dif)
315 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
316 dif, 0, &udp_table, NULL);
317 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
321 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
324 /* do not use the scratch area len for jumbogram: their length execeeds the
325 * scratch area space; note that the IP6CB flags is still in the first
326 * cacheline, so checking for jumbograms is cheap
328 static int udp6_skb_len(struct sk_buff *skb)
330 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
334 * This should be easy, if there is something there we
335 * return it, otherwise we block.
338 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
339 int flags, int *addr_len)
341 struct ipv6_pinfo *np = inet6_sk(sk);
342 struct inet_sock *inet = inet_sk(sk);
344 unsigned int ulen, copied;
345 int off, err, peeking = flags & MSG_PEEK;
346 int is_udplite = IS_UDPLITE(sk);
347 struct udp_mib __percpu *mib;
348 bool checksum_valid = false;
351 if (flags & MSG_ERRQUEUE)
352 return ipv6_recv_error(sk, msg, len, addr_len);
354 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
355 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
358 off = sk_peek_offset(sk, flags);
359 skb = __skb_recv_udp(sk, flags, &off, &err);
363 ulen = udp6_skb_len(skb);
365 if (copied > ulen - off)
367 else if (copied < ulen)
368 msg->msg_flags |= MSG_TRUNC;
370 is_udp4 = (skb->protocol == htons(ETH_P_IP));
371 mib = __UDPX_MIB(sk, is_udp4);
374 * If checksum is needed at all, try to do it while copying the
375 * data. If the data is truncated, or if we only want a partial
376 * coverage checksum (UDP-Lite), do it before the copy.
379 if (copied < ulen || peeking ||
380 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
381 checksum_valid = udp_skb_csum_unnecessary(skb) ||
382 !__udp_lib_checksum_complete(skb);
387 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
388 if (udp_skb_is_linear(skb))
389 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
391 err = skb_copy_datagram_msg(skb, off, msg, copied);
393 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
399 atomic_inc(&sk->sk_drops);
400 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
406 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
408 sock_recv_cmsgs(msg, sk, skb);
410 /* Copy the address. */
412 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
413 sin6->sin6_family = AF_INET6;
414 sin6->sin6_port = udp_hdr(skb)->source;
415 sin6->sin6_flowinfo = 0;
418 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
420 sin6->sin6_scope_id = 0;
422 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
423 sin6->sin6_scope_id =
424 ipv6_iface_scope_id(&sin6->sin6_addr,
427 *addr_len = sizeof(*sin6);
429 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
430 (struct sockaddr *)sin6);
433 if (udp_sk(sk)->gro_enabled)
434 udp_cmsg_recv(msg, sk, skb);
437 ip6_datagram_recv_common_ctl(sk, msg, skb);
440 if (inet->cmsg_flags)
441 ip_cmsg_recv_offset(msg, sk, skb,
442 sizeof(struct udphdr), off);
445 ip6_datagram_recv_specific_ctl(sk, msg, skb);
449 if (flags & MSG_TRUNC)
452 skb_consume_udp(sk, skb, peeking ? -err : err);
456 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
457 udp_skb_destructor)) {
458 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
459 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
463 /* starting over for a new packet, but check if we need to yield */
465 msg->msg_flags &= ~MSG_TRUNC;
469 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
470 void udpv6_encap_enable(void)
472 static_branch_inc(&udpv6_encap_needed_key);
474 EXPORT_SYMBOL(udpv6_encap_enable);
476 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
477 * through error handlers in encapsulations looking for a match.
479 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
480 struct inet6_skb_parm *opt,
481 u8 type, u8 code, int offset, __be32 info)
485 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
486 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
487 u8 type, u8 code, int offset, __be32 info);
488 const struct ip6_tnl_encap_ops *encap;
490 encap = rcu_dereference(ip6tun_encaps[i]);
493 handler = encap->err_handler;
494 if (handler && !handler(skb, opt, type, code, offset, info))
501 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
502 * reversing source and destination port: this will match tunnels that force the
503 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
504 * lwtunnels might actually break this assumption by being configured with
505 * different destination ports on endpoints, in this case we won't be able to
506 * trace ICMP messages back to them.
508 * If this doesn't match any socket, probe tunnels with arbitrary destination
509 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
510 * we've sent packets to won't necessarily match the local destination port.
512 * Then ask the tunnel implementation to match the error against a valid
515 * Return an error if we can't find a match, the socket if we need further
516 * processing, zero otherwise.
518 static struct sock *__udp6_lib_err_encap(struct net *net,
519 const struct ipv6hdr *hdr, int offset,
521 struct udp_table *udptable,
524 struct inet6_skb_parm *opt,
525 u8 type, u8 code, __be32 info)
527 int (*lookup)(struct sock *sk, struct sk_buff *skb);
528 int network_offset, transport_offset;
531 network_offset = skb_network_offset(skb);
532 transport_offset = skb_transport_offset(skb);
534 /* Network header needs to point to the outer IPv6 header inside ICMP */
535 skb_reset_network_header(skb);
537 /* Transport header needs to point to the UDP header */
538 skb_set_transport_header(skb, offset);
543 lookup = READ_ONCE(up->encap_err_lookup);
544 if (lookup && lookup(sk, skb))
550 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
551 &hdr->saddr, uh->dest,
552 inet6_iif(skb), 0, udptable, skb);
556 lookup = READ_ONCE(up->encap_err_lookup);
557 if (!lookup || lookup(sk, skb))
563 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
567 skb_set_transport_header(skb, transport_offset);
568 skb_set_network_header(skb, network_offset);
573 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
574 u8 type, u8 code, int offset, __be32 info,
575 struct udp_table *udptable)
577 struct ipv6_pinfo *np;
578 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
579 const struct in6_addr *saddr = &hdr->saddr;
580 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr;
581 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
586 struct net *net = dev_net(skb->dev);
588 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
589 inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
591 if (!sk || udp_sk(sk)->encap_type) {
592 /* No socket for error: try tunnels before discarding */
593 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
594 sk = __udp6_lib_err_encap(net, hdr, offset, uh,
596 opt, type, code, info);
600 sk = ERR_PTR(-ENOENT);
603 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
611 harderr = icmpv6_err_convert(type, code, &err);
614 if (type == ICMPV6_PKT_TOOBIG) {
615 if (!ip6_sk_accept_pmtu(sk))
617 ip6_sk_update_pmtu(skb, sk, info);
618 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
621 if (type == NDISC_REDIRECT) {
623 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
624 sk->sk_mark, sk->sk_uid);
626 ip6_sk_redirect(skb, sk);
631 /* Tunnels don't have an application socket: don't pass errors back */
633 if (udp_sk(sk)->encap_err_rcv)
634 udp_sk(sk)->encap_err_rcv(sk, skb, offset);
639 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
642 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
651 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
655 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
656 sock_rps_save_rxhash(sk, skb);
657 sk_mark_napi_id(sk, skb);
658 sk_incoming_cpu_update(sk);
660 sk_mark_napi_id_once(sk, skb);
663 rc = __udp_enqueue_schedule_skb(sk, skb);
665 int is_udplite = IS_UDPLITE(sk);
666 enum skb_drop_reason drop_reason;
668 /* Note that an ENOMEM error is charged twice */
670 UDP6_INC_STATS(sock_net(sk),
671 UDP_MIB_RCVBUFERRORS, is_udplite);
672 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
674 UDP6_INC_STATS(sock_net(sk),
675 UDP_MIB_MEMERRORS, is_udplite);
676 drop_reason = SKB_DROP_REASON_PROTO_MEM;
678 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
679 kfree_skb_reason(skb, drop_reason);
686 static __inline__ int udpv6_err(struct sk_buff *skb,
687 struct inet6_skb_parm *opt, u8 type,
688 u8 code, int offset, __be32 info)
690 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
693 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
695 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
696 struct udp_sock *up = udp_sk(sk);
697 int is_udplite = IS_UDPLITE(sk);
699 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
700 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
704 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
705 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
708 * This is an encapsulation socket so pass the skb to
709 * the socket's udp_encap_rcv() hook. Otherwise, just
710 * fall through and pass this up the UDP socket.
711 * up->encap_rcv() returns the following value:
712 * =0 if skb was successfully passed to the encap
713 * handler or was discarded by it.
714 * >0 if skb should be passed on to UDP.
715 * <0 if skb should be resubmitted as proto -N
718 /* if we're overly short, let UDP handle it */
719 encap_rcv = READ_ONCE(up->encap_rcv);
723 /* Verify checksum before giving to encap */
724 if (udp_lib_checksum_complete(skb))
727 ret = encap_rcv(sk, skb);
729 __UDP6_INC_STATS(sock_net(sk),
736 /* FALLTHROUGH -- it's a UDP Packet */
740 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
742 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
744 if (up->pcrlen == 0) { /* full coverage was set */
745 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
746 UDP_SKB_CB(skb)->cscov, skb->len);
749 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
750 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
751 UDP_SKB_CB(skb)->cscov, up->pcrlen);
756 prefetch(&sk->sk_rmem_alloc);
757 if (rcu_access_pointer(sk->sk_filter) &&
758 udp_lib_checksum_complete(skb))
761 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
762 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
766 udp_csum_pull_header(skb);
770 return __udpv6_queue_rcv_skb(sk, skb);
773 drop_reason = SKB_DROP_REASON_UDP_CSUM;
774 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
776 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
777 atomic_inc(&sk->sk_drops);
778 kfree_skb_reason(skb, drop_reason);
782 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
784 struct sk_buff *next, *segs;
787 if (likely(!udp_unexpected_gso(sk, skb)))
788 return udpv6_queue_rcv_one_skb(sk, skb);
790 __skb_push(skb, -skb_mac_offset(skb));
791 segs = udp_rcv_segment(sk, skb, false);
792 skb_list_walk_safe(segs, skb, next) {
793 __skb_pull(skb, skb_transport_offset(skb));
795 udp_post_segment_fix_csum(skb);
796 ret = udpv6_queue_rcv_one_skb(sk, skb);
798 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
804 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
805 __be16 loc_port, const struct in6_addr *loc_addr,
806 __be16 rmt_port, const struct in6_addr *rmt_addr,
807 int dif, int sdif, unsigned short hnum)
809 struct inet_sock *inet = inet_sk(sk);
811 if (!net_eq(sock_net(sk), net))
814 if (udp_sk(sk)->udp_port_hash != hnum ||
815 sk->sk_family != PF_INET6 ||
816 (inet->inet_dport && inet->inet_dport != rmt_port) ||
817 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
818 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
819 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) ||
820 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
821 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
823 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
828 static void udp6_csum_zero_error(struct sk_buff *skb)
830 /* RFC 2460 section 8.1 says that we SHOULD log
831 * this error. Well, it is reasonable.
833 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
834 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
835 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
839 * Note: called only from the BH handler context,
840 * so we don't need to lock the hashes.
842 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
843 const struct in6_addr *saddr, const struct in6_addr *daddr,
844 struct udp_table *udptable, int proto)
846 struct sock *sk, *first = NULL;
847 const struct udphdr *uh = udp_hdr(skb);
848 unsigned short hnum = ntohs(uh->dest);
849 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
850 unsigned int offset = offsetof(typeof(*sk), sk_node);
851 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
852 int dif = inet6_iif(skb);
853 int sdif = inet6_sdif(skb);
854 struct hlist_node *node;
855 struct sk_buff *nskb;
858 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
860 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
862 hslot = &udptable->hash2[hash2];
863 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
866 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
867 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
868 uh->source, saddr, dif, sdif,
871 /* If zero checksum and no_check is not on for
872 * the socket then skip it.
874 if (!uh->check && !udp_sk(sk)->no_check6_rx)
880 nskb = skb_clone(skb, GFP_ATOMIC);
881 if (unlikely(!nskb)) {
882 atomic_inc(&sk->sk_drops);
883 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
885 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
890 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
894 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
895 if (use_hash2 && hash2 != hash2_any) {
901 if (udpv6_queue_rcv_skb(first, skb) > 0)
905 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
906 proto == IPPROTO_UDPLITE);
911 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
913 if (udp_sk_rx_dst_set(sk, dst)) {
914 const struct rt6_info *rt = (const struct rt6_info *)dst;
916 sk->sk_rx_dst_cookie = rt6_get_cookie(rt);
920 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
921 * return code conversion for ip layer consumption
923 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
928 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
929 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
931 ret = udpv6_queue_rcv_skb(sk, skb);
933 /* a return value > 0 means to resubmit the input */
939 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
942 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
943 const struct in6_addr *saddr, *daddr;
944 struct net *net = dev_net(skb->dev);
950 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
953 saddr = &ipv6_hdr(skb)->saddr;
954 daddr = &ipv6_hdr(skb)->daddr;
957 ulen = ntohs(uh->len);
961 if (proto == IPPROTO_UDP) {
962 /* UDP validates ulen. */
964 /* Check for jumbo payload */
968 if (ulen < sizeof(*uh))
971 if (ulen < skb->len) {
972 if (pskb_trim_rcsum(skb, ulen))
974 saddr = &ipv6_hdr(skb)->saddr;
975 daddr = &ipv6_hdr(skb)->daddr;
980 if (udp6_csum_init(skb, uh, proto))
983 /* Check if the socket is already available, e.g. due to early demux */
984 sk = skb_steal_sock(skb, &refcounted);
986 struct dst_entry *dst = skb_dst(skb);
989 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
990 udp6_sk_rx_dst_set(sk, dst);
992 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
995 goto report_csum_error;
998 ret = udp6_unicast_rcv_skb(sk, skb, uh);
1005 * Multicast receive code
1007 if (ipv6_addr_is_multicast(daddr))
1008 return __udp6_lib_mcast_deliver(net, skb,
1009 saddr, daddr, udptable, proto);
1012 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1014 if (!uh->check && !udp_sk(sk)->no_check6_rx)
1015 goto report_csum_error;
1016 return udp6_unicast_rcv_skb(sk, skb, uh);
1019 reason = SKB_DROP_REASON_NO_SOCKET;
1022 goto report_csum_error;
1024 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1027 if (udp_lib_checksum_complete(skb))
1030 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1031 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
1033 kfree_skb_reason(skb, reason);
1037 if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1038 reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1039 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
1040 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1041 saddr, ntohs(uh->source),
1043 daddr, ntohs(uh->dest));
1047 udp6_csum_zero_error(skb);
1049 if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1050 reason = SKB_DROP_REASON_UDP_CSUM;
1051 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1053 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1054 kfree_skb_reason(skb, reason);
1059 static struct sock *__udp6_lib_demux_lookup(struct net *net,
1060 __be16 loc_port, const struct in6_addr *loc_addr,
1061 __be16 rmt_port, const struct in6_addr *rmt_addr,
1064 unsigned short hnum = ntohs(loc_port);
1065 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1066 unsigned int slot2 = hash2 & udp_table.mask;
1067 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1068 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1071 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1072 if (sk->sk_state == TCP_ESTABLISHED &&
1073 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif))
1075 /* Only check first socket in chain */
1081 void udp_v6_early_demux(struct sk_buff *skb)
1083 struct net *net = dev_net(skb->dev);
1084 const struct udphdr *uh;
1086 struct dst_entry *dst;
1087 int dif = skb->dev->ifindex;
1088 int sdif = inet6_sdif(skb);
1090 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1091 sizeof(struct udphdr)))
1096 if (skb->pkt_type == PACKET_HOST)
1097 sk = __udp6_lib_demux_lookup(net, uh->dest,
1098 &ipv6_hdr(skb)->daddr,
1099 uh->source, &ipv6_hdr(skb)->saddr,
1104 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
1108 skb->destructor = sock_efree;
1109 dst = rcu_dereference(sk->sk_rx_dst);
1112 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1114 /* set noref for now.
1115 * any place which wants to hold dst has to call
1118 skb_dst_set_noref(skb, dst);
1122 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1124 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1128 * Throw away all pending data and cancel the corking. Socket is locked.
1130 static void udp_v6_flush_pending_frames(struct sock *sk)
1132 struct udp_sock *up = udp_sk(sk);
1134 if (up->pending == AF_INET)
1135 udp_flush_pending_frames(sk);
1136 else if (up->pending) {
1139 ip6_flush_pending_frames(sk);
1143 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1146 if (addr_len < offsetofend(struct sockaddr, sa_family))
1148 /* The following checks are replicated from __ip6_datagram_connect()
1149 * and intended to prevent BPF program called below from accessing
1150 * bytes that are out of the bound specified by user in addr_len.
1152 if (uaddr->sa_family == AF_INET) {
1153 if (ipv6_only_sock(sk))
1154 return -EAFNOSUPPORT;
1155 return udp_pre_connect(sk, uaddr, addr_len);
1158 if (addr_len < SIN6_LEN_RFC2133)
1161 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr);
1165 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1166 * @sk: socket we are sending on
1167 * @skb: sk_buff containing the filled-in UDP header
1168 * (checksum field must be zeroed out)
1169 * @saddr: source address
1170 * @daddr: destination address
1171 * @len: length of packet
1173 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1174 const struct in6_addr *saddr,
1175 const struct in6_addr *daddr, int len)
1177 unsigned int offset;
1178 struct udphdr *uh = udp_hdr(skb);
1179 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1183 /* Only one fragment on the socket. */
1184 skb->csum_start = skb_transport_header(skb) - skb->head;
1185 skb->csum_offset = offsetof(struct udphdr, check);
1186 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1189 * HW-checksum won't work as there are two or more
1190 * fragments on the socket so that all csums of sk_buffs
1191 * should be together
1193 offset = skb_transport_offset(skb);
1194 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1197 skb->ip_summed = CHECKSUM_NONE;
1200 csum = csum_add(csum, frags->csum);
1201 } while ((frags = frags->next));
1203 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1206 uh->check = CSUM_MANGLED_0;
1214 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1215 struct inet_cork *cork)
1217 struct sock *sk = skb->sk;
1220 int is_udplite = IS_UDPLITE(sk);
1222 int offset = skb_transport_offset(skb);
1223 int len = skb->len - offset;
1224 int datalen = len - sizeof(*uh);
1227 * Create a UDP header
1230 uh->source = fl6->fl6_sport;
1231 uh->dest = fl6->fl6_dport;
1232 uh->len = htons(len);
1235 if (cork->gso_size) {
1236 const int hlen = skb_network_header_len(skb) +
1237 sizeof(struct udphdr);
1239 if (hlen + cork->gso_size > cork->fragsize) {
1243 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1247 if (udp_sk(sk)->no_check6_tx) {
1251 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1252 dst_xfrm(skb_dst(skb))) {
1257 if (datalen > cork->gso_size) {
1258 skb_shinfo(skb)->gso_size = cork->gso_size;
1259 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1260 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1267 csum = udplite_csum(skb);
1268 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1269 skb->ip_summed = CHECKSUM_NONE;
1271 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1273 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1276 csum = udp_csum(skb);
1278 /* add protocol-dependent pseudo-header */
1279 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1280 len, fl6->flowi6_proto, csum);
1282 uh->check = CSUM_MANGLED_0;
1285 err = ip6_send_skb(skb);
1287 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1288 UDP6_INC_STATS(sock_net(sk),
1289 UDP_MIB_SNDBUFERRORS, is_udplite);
1293 UDP6_INC_STATS(sock_net(sk),
1294 UDP_MIB_OUTDATAGRAMS, is_udplite);
1299 static int udp_v6_push_pending_frames(struct sock *sk)
1301 struct sk_buff *skb;
1302 struct udp_sock *up = udp_sk(sk);
1305 if (up->pending == AF_INET)
1306 return udp_push_pending_frames(sk);
1308 skb = ip6_finish_skb(sk);
1312 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6,
1313 &inet_sk(sk)->cork.base);
1320 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1322 struct ipv6_txoptions opt_space;
1323 struct udp_sock *up = udp_sk(sk);
1324 struct inet_sock *inet = inet_sk(sk);
1325 struct ipv6_pinfo *np = inet6_sk(sk);
1326 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1327 struct in6_addr *daddr, *final_p, final;
1328 struct ipv6_txoptions *opt = NULL;
1329 struct ipv6_txoptions *opt_to_free = NULL;
1330 struct ip6_flowlabel *flowlabel = NULL;
1331 struct inet_cork_full cork;
1332 struct flowi6 *fl6 = &cork.fl.u.ip6;
1333 struct dst_entry *dst;
1334 struct ipcm6_cookie ipc6;
1335 int addr_len = msg->msg_namelen;
1336 bool connected = false;
1338 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1340 int is_udplite = IS_UDPLITE(sk);
1341 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1344 ipc6.gso_size = READ_ONCE(up->gso_size);
1345 ipc6.sockc.tsflags = sk->sk_tsflags;
1346 ipc6.sockc.mark = sk->sk_mark;
1348 /* destination address check */
1350 if (addr_len < offsetof(struct sockaddr, sa_data))
1353 switch (sin6->sin6_family) {
1355 if (addr_len < SIN6_LEN_RFC2133)
1357 daddr = &sin6->sin6_addr;
1358 if (ipv6_addr_any(daddr) &&
1359 ipv6_addr_v4mapped(&np->saddr))
1360 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1364 goto do_udp_sendmsg;
1366 msg->msg_name = sin6 = NULL;
1367 msg->msg_namelen = addr_len = 0;
1373 } else if (!up->pending) {
1374 if (sk->sk_state != TCP_ESTABLISHED)
1375 return -EDESTADDRREQ;
1376 daddr = &sk->sk_v6_daddr;
1381 if (ipv6_addr_v4mapped(daddr)) {
1382 struct sockaddr_in sin;
1383 sin.sin_family = AF_INET;
1384 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1385 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1386 msg->msg_name = &sin;
1387 msg->msg_namelen = sizeof(sin);
1389 if (ipv6_only_sock(sk))
1390 return -ENETUNREACH;
1391 return udp_sendmsg(sk, msg, len);
1395 /* Rough check on arithmetic overflow,
1396 better check is made in ip6_append_data().
1398 if (len > INT_MAX - sizeof(struct udphdr))
1401 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1403 if (up->pending == AF_INET)
1404 return udp_sendmsg(sk, msg, len);
1406 * There are pending frames.
1407 * The socket lock must be held while it's corked.
1410 if (likely(up->pending)) {
1411 if (unlikely(up->pending != AF_INET6)) {
1413 return -EAFNOSUPPORT;
1416 goto do_append_data;
1420 ulen += sizeof(struct udphdr);
1422 memset(fl6, 0, sizeof(*fl6));
1425 if (sin6->sin6_port == 0)
1428 fl6->fl6_dport = sin6->sin6_port;
1429 daddr = &sin6->sin6_addr;
1432 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1433 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) {
1434 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1435 if (IS_ERR(flowlabel))
1441 * Otherwise it will be difficult to maintain
1444 if (sk->sk_state == TCP_ESTABLISHED &&
1445 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1446 daddr = &sk->sk_v6_daddr;
1448 if (addr_len >= sizeof(struct sockaddr_in6) &&
1449 sin6->sin6_scope_id &&
1450 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1451 fl6->flowi6_oif = sin6->sin6_scope_id;
1453 if (sk->sk_state != TCP_ESTABLISHED)
1454 return -EDESTADDRREQ;
1456 fl6->fl6_dport = inet->inet_dport;
1457 daddr = &sk->sk_v6_daddr;
1458 fl6->flowlabel = np->flow_label;
1462 if (!fl6->flowi6_oif)
1463 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if);
1465 if (!fl6->flowi6_oif)
1466 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1468 fl6->flowi6_uid = sk->sk_uid;
1470 if (msg->msg_controllen) {
1472 memset(opt, 0, sizeof(struct ipv6_txoptions));
1473 opt->tot_len = sizeof(*opt);
1476 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1478 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6,
1481 fl6_sock_release(flowlabel);
1484 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1485 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1486 if (IS_ERR(flowlabel))
1489 if (!(opt->opt_nflen|opt->opt_flen))
1494 opt = txopt_get(np);
1498 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1499 opt = ipv6_fixup_options(&opt_space, opt);
1502 fl6->flowi6_proto = sk->sk_protocol;
1503 fl6->flowi6_mark = ipc6.sockc.mark;
1504 fl6->daddr = *daddr;
1505 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr))
1506 fl6->saddr = np->saddr;
1507 fl6->fl6_sport = inet->inet_sport;
1509 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) {
1510 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1511 (struct sockaddr *)sin6,
1516 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1517 /* BPF program rewrote IPv6-only by IPv4-mapped
1518 * IPv6. It's currently unsupported.
1523 if (sin6->sin6_port == 0) {
1524 /* BPF program set invalid port. Reject it. */
1528 fl6->fl6_dport = sin6->sin6_port;
1529 fl6->daddr = sin6->sin6_addr;
1533 if (ipv6_addr_any(&fl6->daddr))
1534 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1536 final_p = fl6_update_dst(fl6, opt, &final);
1540 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) {
1541 fl6->flowi6_oif = np->mcast_oif;
1543 } else if (!fl6->flowi6_oif)
1544 fl6->flowi6_oif = np->ucast_oif;
1546 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6));
1548 if (ipc6.tclass < 0)
1549 ipc6.tclass = np->tclass;
1551 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel);
1553 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected);
1560 if (ipc6.hlimit < 0)
1561 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst);
1563 if (msg->msg_flags&MSG_CONFIRM)
1567 /* Lockless fast path for the non-corking case */
1569 struct sk_buff *skb;
1571 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1572 sizeof(struct udphdr), &ipc6,
1573 (struct rt6_info *)dst,
1574 msg->msg_flags, &cork);
1576 if (!IS_ERR_OR_NULL(skb))
1577 err = udp_v6_send_skb(skb, fl6, &cork.base);
1578 /* ip6_make_skb steals dst reference */
1583 if (unlikely(up->pending)) {
1584 /* The socket is already corked while preparing it. */
1585 /* ... which is an evident application bug. --ANK */
1588 net_dbg_ratelimited("udp cork app bug 2\n");
1593 up->pending = AF_INET6;
1596 if (ipc6.dontfrag < 0)
1597 ipc6.dontfrag = np->dontfrag;
1599 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1600 &ipc6, fl6, (struct rt6_info *)dst,
1601 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1603 udp_v6_flush_pending_frames(sk);
1605 err = udp_v6_push_pending_frames(sk);
1606 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1610 err = np->recverr ? net_xmit_errno(err) : 0;
1616 fl6_sock_release(flowlabel);
1617 txopt_put(opt_to_free);
1621 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1622 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1623 * we don't have a good statistic (IpOutDiscards but it can be too many
1624 * things). We could add another new stat but at least for now that
1625 * seems like overkill.
1627 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1628 UDP6_INC_STATS(sock_net(sk),
1629 UDP_MIB_SNDBUFERRORS, is_udplite);
1634 if (msg->msg_flags & MSG_PROBE)
1635 dst_confirm_neigh(dst, &fl6->daddr);
1636 if (!(msg->msg_flags&MSG_PROBE) || len)
1637 goto back_from_confirm;
1642 void udpv6_destroy_sock(struct sock *sk)
1644 struct udp_sock *up = udp_sk(sk);
1647 /* protects from races with udp_abort() */
1648 sock_set_flag(sk, SOCK_DEAD);
1649 udp_v6_flush_pending_frames(sk);
1652 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1653 if (up->encap_type) {
1654 void (*encap_destroy)(struct sock *sk);
1655 encap_destroy = READ_ONCE(up->encap_destroy);
1659 if (up->encap_enabled) {
1660 static_branch_dec(&udpv6_encap_needed_key);
1661 udp_encap_disable();
1665 inet6_destroy_sock(sk);
1669 * Socket option code for UDP
1671 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1672 unsigned int optlen)
1674 if (level == SOL_UDP || level == SOL_UDPLITE)
1675 return udp_lib_setsockopt(sk, level, optname,
1677 udp_v6_push_pending_frames);
1678 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1681 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1682 char __user *optval, int __user *optlen)
1684 if (level == SOL_UDP || level == SOL_UDPLITE)
1685 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1686 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1689 static const struct inet6_protocol udpv6_protocol = {
1690 .handler = udpv6_rcv,
1691 .err_handler = udpv6_err,
1692 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1695 /* ------------------------------------------------------------------------ */
1696 #ifdef CONFIG_PROC_FS
1697 int udp6_seq_show(struct seq_file *seq, void *v)
1699 if (v == SEQ_START_TOKEN) {
1700 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1702 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1703 struct inet_sock *inet = inet_sk(v);
1704 __u16 srcp = ntohs(inet->inet_sport);
1705 __u16 destp = ntohs(inet->inet_dport);
1706 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1707 udp_rqueue_get(v), bucket);
1712 const struct seq_operations udp6_seq_ops = {
1713 .start = udp_seq_start,
1714 .next = udp_seq_next,
1715 .stop = udp_seq_stop,
1716 .show = udp6_seq_show,
1718 EXPORT_SYMBOL(udp6_seq_ops);
1720 static struct udp_seq_afinfo udp6_seq_afinfo = {
1722 .udp_table = &udp_table,
1725 int __net_init udp6_proc_init(struct net *net)
1727 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1728 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1733 void udp6_proc_exit(struct net *net)
1735 remove_proc_entry("udp6", net->proc_net);
1737 #endif /* CONFIG_PROC_FS */
1739 /* ------------------------------------------------------------------------ */
1741 struct proto udpv6_prot = {
1743 .owner = THIS_MODULE,
1744 .close = udp_lib_close,
1745 .pre_connect = udpv6_pre_connect,
1746 .connect = ip6_datagram_connect,
1747 .disconnect = udp_disconnect,
1749 .init = udpv6_init_sock,
1750 .destroy = udpv6_destroy_sock,
1751 .setsockopt = udpv6_setsockopt,
1752 .getsockopt = udpv6_getsockopt,
1753 .sendmsg = udpv6_sendmsg,
1754 .recvmsg = udpv6_recvmsg,
1755 .release_cb = ip6_datagram_release_cb,
1756 .hash = udp_lib_hash,
1757 .unhash = udp_lib_unhash,
1758 .rehash = udp_v6_rehash,
1759 .get_port = udp_v6_get_port,
1760 .put_port = udp_lib_unhash,
1761 #ifdef CONFIG_BPF_SYSCALL
1762 .psock_update_sk_prot = udp_bpf_update_proto,
1765 .memory_allocated = &udp_memory_allocated,
1766 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
1768 .sysctl_mem = sysctl_udp_mem,
1769 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1770 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1771 .obj_size = sizeof(struct udp6_sock),
1772 .h.udp_table = &udp_table,
1773 .diag_destroy = udp_abort,
1776 static struct inet_protosw udpv6_protosw = {
1778 .protocol = IPPROTO_UDP,
1779 .prot = &udpv6_prot,
1780 .ops = &inet6_dgram_ops,
1781 .flags = INET_PROTOSW_PERMANENT,
1784 int __init udpv6_init(void)
1788 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1792 ret = inet6_register_protosw(&udpv6_protosw);
1794 goto out_udpv6_protocol;
1799 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1803 void udpv6_exit(void)
1805 inet6_unregister_protosw(&udpv6_protosw);
1806 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);